[PATCH] D69989: Assume ieee behavior without denormal-fp-math attribute

[Sanjay Patel via Phabricator via llvm-commits]

spatel added inline comments.


================
Comment at: llvm/test/CodeGen/X86/sqrt-fastmath-tune.ll:101
 attributes #1 = { nounwind readnone }
+attributes #2 = { "denormal-fp-math"="ieee,preseve-sign" "unsafe-fp-math"="true" }
----------------
arsenm wrote:
> spatel wrote:
> > arsenm wrote:
> > > spatel wrote:
> > > > Does the "preseve" typo affect anything?
> > > These aren't hitting the expansion, so I guess it doesn't. I assume this has something to do with not using reciprocal-estimates? 
> > Yes, this file isn't well-specified about what it's trying to test for. I updated it here:
> > rG6071fc57a45f
> > 
> > That should remove it from this patch?
> > 
> > Independent of that - should we have an assert or warning that would catch typos in the attribute string?
> I remember implementing a verifier check for this before ,but I can't seem to find the patch
It would be good to revive that patch for safety. But this patch doesn't need to be gated by that improvement (IIUC, these diffs disappear if you rebase?). 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D69989/new/

https://reviews.llvm.org/D69989